|
|||||
|
|
| 生成模型学习的遥感影像半监督分类 | |
| 引用本文： | 任广波,张杰,马毅,郑荣儿.生成模型学习的遥感影像半监督分类[J].遥感学报,2010,14(6):1097-1110. |
| 作者姓名： | 任广波 张杰 马毅 郑荣儿 |
| 作者单位： | 1. 中国海洋大学,光学光电子实验室,山东,青岛,266100;国家海洋局第一海洋研究所,山东,青岛,266061 2. 国家海洋局第一海洋研究所,山东,青岛,266061;国家海洋局第一海洋研究所,海洋环境和数值模拟国家海洋局重点实验室,山东,青岛,266061 3. 中国海洋大学,光学光电子实验室,山东,青岛,266100 |
| 基金项目： | 国家自然科学基金青年基金(编号: 40906094); 中国近海海洋综合调查与评价专项(编号: 908-01-WY08) |
| 摘 要： | 以生成模型最大似然估计为例,引入结合已标记样本和未标记样本的半监督分类方法来解决遥感影像分类中的小样本问题,应用已有的少量已标记样本初始化一个分类器,结合大量未标记样本,通过递归计算的方式对分类器进行优化,直到包含所有样本的似然函数收敛到局部极大值。通过分析遥感影像待分类类别与影像中地物类型固有特征之间的关系,设计两个在不同生成模型假设下的分类实验。结果表明,未标记样本的参与可在很大程度上提高小样本条件下的影像分类精度,但两种样本的数量应保持一个适当的比例。最后通过与在解决小样本分类问题方面有独特优势的SVM方法的分类比较,发现在小样本情况下,本文方法具有更好的应用潜力。 |
| 关 键 词： | 遥感分类 半监督学习 EM 算法 |
| 收稿时间： | 2009/10/13 0:00:00 |
| 修稿时间： | 5/7/2010 12:00:00 AM |
Generative model based semi-supervised learning method of\nremote sensing image classification |
|
| REN Guangbo,ZHANG Jie,MA Yi,ZHENG Rong'er.Generative model based semi-supervised learning method of\nremote sensing image classification[J].Journal of Remote Sensing,2010,14(6):1097-1110. | |
| Authors: | REN Guangbo ZHANG Jie MA Yi ZHENG Rong'er |
| Institution: | Optics & OptiElectronics Laboratory, Ocean University of China, Shandong Qingdao 266100, China;First Institute of Oceanography, State Oceanic Administration(SOA), Shandong Qingdao 266061, China;First Institute of Oceanography, State Oceanic Administration(SOA), Shandong Qingdao 266061, China;Key Laboratory of Marine Science and Numerical Modeling, SOA, Shandong Qingdao 266061, China;First Institute of Oceanography, State Oceanic Administration(SOA), Shandong Qingdao 266061, China;Key Laboratory of Marine Science and Numerical Modeling, SOA, Shandong Qingdao 266061, China;Optics & OptiElectronics Laboratory, Ocean University of China, Shandong Qingdao 266100, China; |
| Abstract: | This paper proposes a generative model based semi-supervised learning method of remote sensing image classification, which makes use of both the labeled and unlabeled samples to handle the insufficient labeled training samples problems. We first train an original classifier by the small number of labeled samples alone. Then we re-train it by both the labeled and a large amount of unlabeled samples. This process is iterated until the likelihood function of all the samples are converged to the local maxima. Through the designed experiments of the two different mixture models, It is found that the unlabeled samples help us to get the method to enhance the classification performance to a large extent on condition, which the ratio of the unlabeled samples to the labeled ones must be appropriate. Thus, we have also compared the method by using the state-of-the-art support vector machines (SVMs) with the same labeled samples, of which results show that our method works better. |
| Keywords: | remote sensing classification semi-supervised learning Expectation Maximum (EM) algorithm |
| 本文献已被 CNKI 万方数据 等数据库收录！ | |
| 点击此处可从《遥感学报》浏览原始摘要信息 | |
| 点击此处可从《遥感学报》下载免费的PDF全文 | |